A collection of AI experts formerly employed by leading companies such as Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs has revealed their plans to establish a new venture called Trajectory. Announced on Wednesday, the startup's goal is to enable organizations to enhance their AI products through continuous training based on real-world user interactions.
Trajectory is focused on creating a platform that supports ongoing learning for AI, a capability that has been recognized by researchers as a significant hurdle to advancements in the field. Industry giants like OpenAI, Google, and Anthropic have successfully developed increasingly sophisticated AI models for tasks like coding, mathematics, and scientific applications. Nonetheless, these systems cease to evolve once their training concludes. Although there have been recent strides in continual learning, tech firms still face challenges in producing AI solutions that can learn and adapt from their mistakes in real-time. During December 2025's NeurIPS, a prominent annual AI research conference, Richard Sutton—recipient of the Turing Award—stressed that continual learning is crucial for developing superintelligent agents.
Trajectory has secured $15 million in seed funding, achieving a post-money valuation of $115 million, with the investment led by Conviction and contributions from Bessemer Venture Partners, Radical VC, and BoxGroup. Notable individual investors include Google DeepMind's chief scientist, Jeff Dean, and the renowned "godmother of AI," Stanford’s Fei-Fei Li, who is also CEO of World Labs.
Leading the company as CEO and cofounder is Ronak Malde, who previously engaged in AI research at Windsurf before moving to Google DeepMind after the coding startup was acquired in a $2.4 billion deal last year. The founding team also includes Arjun Karanam, an ex-AI researcher at Apple known for contributions to the Vision Pro, and Michael Elabd, who was part of Google DeepMind's robotics division.
In an interview with WIRED, Malde highlighted that some pioneering AI coding solutions, such as Cursor, are already implementing early forms of continual learning. These products leverage actual user data to refine their models post-training and continuously push updates. Malde believes this iterative approach has significantly contributed to the rapid growth of AI coding assistance tools and has motivated major labs to create their own advanced coding applications. With Trajectory, Malde and his team of 11 researchers and engineers aim to extend similar continuous improvement strategies to AI tools beyond the coding realm.
"Currently, the most advanced AIs remain static," Malde explained. "The model you relied on yesterday will make the same errors today. A handful of companies are edging toward continual learning. Our mission is to build a platform that empowers every organization to achieve this capability."
